X3DFritz vs Kasparov, what conclusions?
Tommy
I just wanted to share with you my thoughts...
If Fritz is going to win game 4, he will have won the whole challenge....
but, would this make Fritz stronger than Kasparov? I do not think so.
I believe whatever happens, whatever the results will be, computers are
still a bit weaker than top grandmasters.
Of course grandmasters are humans and they can make mistakes.... but if a
grandmaster makes no major mistakes, then they are still a bit better than
computers.
The result of the last match (tomorrow) will depend on Kasparov... if he
will make a mistake Frtiz can win, otherwise Fritz can at the most draw.
(I believe the game will probably be a draw.....)
Computers are still weak at long-term-strategies (on the other hands are
unbeatable at tactics!).
Tommy
Mhoulsby
>Date: 17/11/03 20:23 GMT Standard Time
>Message-id: <bpbao4$55m$1...@titan.btinternet.com>
Yes, I could not have put it better.
Mark
henri Arsenault
the result, the conclusion will be based on a single game, and therefore
of little statistical value. Of course if Kasparov wins, some will
conclude that it proves that humans are stronger than computers, and if
Fritz wins, there will be some who conclude that it proves that
computers are now stronger than humans.
Neither conclusion will be warranted, no more than any conclusion could
be derived from a single win between two strong grandmasters.
As a matter of fact, a 4-game match between approximately evenly matched
opponents is statistically insignificant, unless one side won at least 3
of the games - and even then. have we forgotten that Fischer was losing
by two games early in the match with Spassky? Did that make Spassky the
better player? There are plenty of similar examples in chess History,
involving players like Lasker and Capablanca as well as other strong
players.
The real interest of the match is not to determine whether Fritz or
kasparov is stronger, but who will "prevail" - be it by only half a
point. Can Kasparov beat Fritz by locking up the position with Black as
he did with White? If not can he stand the pressure? Will Fritz come up
with unexpected good moves? Will the Grandmaster analysts make fools of
themselves? Will kasparov be a sore loser? Will Fritz sales increase if
Fritz wins? These are the questions that make it worthwhile to follow
the match.
Henri
Chuck in Minot
monsters teeth from about 10 seconds after Kaspy got up from his seat after
the resignation. The Fritz he will be playing tomorrow will not have its
king pawn restriction value set so high, and along with its normal search
for the best move, it will also be checking for locked positions and blocked
pawn structures, and false guarding, (Kaspy's genius little rook fake) as
well as best moves, and play accordingly..
Now you know that Kaspy is well aware of these, and you can bet that he has
yet another plan, and will be ready for these changes, and will win. Kaspy
will have to play in a way to ensure that he keeps poking his fingers in the
monster's tactical eye, and use moves to throw off its calculations (Still
in awe of his fake rook move to fool the monster about the pawn wall) He
now knows how the monster thinks, and will play to lead it on. If he can
fool the monster into thinking one thing, while planning another, and send
the beast back home again, we should all be able to look forward to a
savings of at least $15.00 off the next release of X3D Fritz. (GRIN)
Good luck Kaspy!!!
Dick Schneiders
>kasparov is stronger, but who will "prevail" - be it by only half a
>point.
Uhh - Neither will prevail by "only half a point". It can never happen.
Robert Hyatt
Not in a match, for sure. :)
--
Robert M. Hyatt, Ph.D. Computer and Information Sciences
hy...@uab.edu University of Alabama at Birmingham
(205) 934-2213 136A Campbell Hall
(205) 934-5473 FAX Birmingham, AL 35294-1170
Robert Hyatt
> I know one thing, and that is Team Fritz have been very busy sharpening the
> monsters teeth from about 10 seconds after Kaspy got up from his seat after
> the resignation. The Fritz he will be playing tomorrow will not have its
> king pawn restriction value set so high, and along with its normal search
> for the best move, it will also be checking for locked positions and blocked
> pawn structures, and false guarding, (Kaspy's genius little rook fake) as
> well as best moves, and play accordingly..
You greatly over-simplify the work required. To add the blocked pawn
stuff would take weeks to write and months to test and tune. Not 2 days.
> Now you know that Kaspy is well aware of these, and you can bet that he has
> yet another plan, and will be ready for these changes, and will win. Kaspy
> will have to play in a way to ensure that he keeps poking his fingers in the
> monster's tactical eye, and use moves to throw off its calculations (Still
> in awe of his fake rook move to fool the monster about the pawn wall) He
> now knows how the monster thinks, and will play to lead it on. If he can
> fool the monster into thinking one thing, while planning another, and send
> the beast back home again, we should all be able to look forward to a
> savings of at least $15.00 off the next release of X3D Fritz. (GRIN)
> Good luck Kaspy!!!
Chuck in Minot
more than you.
If this was Crafty, and upon examining the games played so far, what
adjustments would you attempt to be making?
Also would you accept a chance to let Crafty play Kaspy?
==========================
David Richerby
> Whatever the result is tomorrow, and whatever conclusion is drawn from
> the result, the conclusion will be based on a single game, and therefore
> of little statistical value.
No, the conclusion will be based on four games. To claim that it is based
on just one is to claim that the result of the first three games does not
influence the outcome of the match.
Dave.
--
David Richerby Impossible Gigantic Whisky (TM):
www.chiark.greenend.org.uk/~davidr/ it's like a single-malt whisky but
it's huge and it can't exist!
Robert Hyatt
> more than you.
> If this was Crafty, and upon examining the games played so far, what
> adjustments would you attempt to be making?
> Also would you accept a chance to let Crafty play Kaspy?
The one thing Crafty does pretty well is avoid the positions such as
those in game 3. IE on ICC the "draw masters" there try just such
tactics all the time to draw the computers (that are usually higher-
rated on ICC than the GM players that challenge them.)
Crafty would never play the sequence of moves as played by Deep Fritz
on Sunday. But that is not to say that it wouldn't have lost in a worse
way for all I know. And just because Crafty is very adept at avoiding
closed positions, doesn't mean it would do well against a player of
Kasparov's skill. IE if a program tries _too_ hard to avoid blocked
positions, the opponent can use that to force it to weaken things as it
avoids blocking pawns. It might end up with pawn weaknesses that would
lose the game, where the blocked position might just end in a draw.
This can be a double-edged evaluation idea. Yes open positions generally
favor the computer, but not if it has to totally wreck (say) its pawn
structure to open things up. Or if it does something like give the
opponent an outpost knight at e6 that can never be driven away, just so
it didn't have to deal with a 4-pawn blocked chain like game 3.
I'd certainly be willing to play him, and there is much better hardware
around than what Fritz is running on. IE I have seen near 7M nodes per
second on a quad opteron running at only 1.8ghz. 2.2ghz opterons are
out, and 8-way boxes are pretty common as well (although not cheap).
That could turn into well beyond 15M nodes per second, which would
be pretty dang fast...
Enough to beat Kasparov? If he plays like in game 2, absolutely. If
he plays like in game three, who knows. :)
Dick Schneiders
>>>The real interest of the match is not to determine whether Fritz or
>>>kasparov is stronger, but who will "prevail" - be it by only half a
>>>point.
I replied:
>> Uhh - Neither will prevail by "only half a point". It can never happen.
Dr. Bob confirmed:
>Not in a match, for sure. :)
Nor in a single game or anything else where there is a full point doled out for
each game!
Dick Schneiders
Bernardz
dav...@chiark.greenend.org.uk says...
> henri Arsenault <ars...@nospamhere.phy.ulaval.ca> wrote:
> > Whatever the result is tomorrow, and whatever conclusion is drawn from
> > the result, the conclusion will be based on a single game, and therefore
> > of little statistical value.
>
> No, the conclusion will be based on four games. To claim that it is based
> on just one is to claim that the result of the first three games does not
> influence the outcome of the match.
>
>
Disagree. You have to go by the score. If Fritz wins tomorrow, it won
two games, drew one and lost one. So we would have to conclude that it
was better.
--
A fool can beat any foolproof system.
14th saying of Bernard
David Richerby
> dav...@chiark.greenend.org.uk says...
>> No, the conclusion will be based on four games. To claim that it is
>> based on just one is to claim that the result of the first three
>> games does not influence the outcome of the match.
>
> Disagree. You have to go by the score.
You don't appear to be disagreeing with anything I said; in fact, you seem
to be agreeing with me. Please try to follow up to the right post.
> If Fritz wins tomorrow, it won two games, drew one and lost one. So we
> would have to conclude that it was better.
You really can't conclude much from a four-game match. It's too
susceptible to external factors. Imagine Kasparov turns up to today's
game with a blistering headache and loses: in a twenty-game match, losing
one game because of a headache would have very little effect; in a
four-game match, there's no time to come back. So, if Fritz wins this
match, you have a little bit of evidence that Fritz plays better chess,
but not very much. You need a longer match to be able to say with
reasonable certainty that one player is better than another.
Dave.
--
David Richerby Poisonous Smokes (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ pack of cigarettes but it'll kill you
in seconds!
Leonardo Ljubicic
"Robert Hyatt" <hy...@crafty.cis.uab.edu> wrote in message
news:bpbqmj$sbe$1...@juniper.cis.uab.edu...
> Chuck in Minot <no...@earthmail.com> wrote:
> > Yes, very true, and nobody would know or understand
> > more than you.
>
> > If this was Crafty, and upon examining the games played so far, what
> > adjustments would you attempt to be making?
>
> > Also would you accept a chance to let Crafty play Kaspy?
>
> The one thing Crafty does pretty well is avoid the positions such as
> those in game 3. IE on ICC the "draw masters" there try just such
> tactics all the time to draw the computers (that are usually higher-
> rated on ICC than the GM players that challenge them.)
>
> Crafty would never play the sequence of moves as played by Deep Fritz
> on Sunday. But that is not to say that it wouldn't have lost in a worse
> way for all I know. And just because Crafty is very adept at avoiding
> closed positions, doesn't mean it would do well against a player of
> Kasparov's skill. IE if a program tries _too_ hard to avoid blocked
> positions, the opponent can use that to force it to weaken things as it
> avoids blocking pawns. It might end up with pawn weaknesses that would
> lose the game, where the blocked position might just end in a draw.
I was hoping someone will say this; is it only my impression that quite a
few of the today's top engines wouldn't behave so stupid as Fritz did in
game 3? Can anyone test Shredder 7.04, Junior 8 and Hiarcs 9 in critical
positions? There is no doubt Fritz made significant progress over the years,
but it still (from what I witness) has very little chess "knowledge", as
compared to other top engines. I recall Fritz got reputation "fast but
stupid".
Leonardo
henri Arsenault
David Richerby <dav...@chiark.greenend.org.uk> wrote:
> henri Arsenault <ars...@nospamhere.phy.ulaval.ca> wrote:
> > Whatever the result is tomorrow, and whatever conclusion is drawn from
> > the result, the conclusion will be based on a single game, and therefore
> > of little statistical value.
>
> No, the conclusion will be based on four games. To claim that it is based
> on just one is to claim that the result of the first three games does not
> influence the outcome of the match.
>
No, it is to conclude that since the score is even after 3 games, these
games have proven nothing about relative strength, and they will not
have proven anything more by adding a fourth game.
Henri
henri Arsenault
dicks...@aol.com (Dick Schneiders) wrote:
> Dr. Bob confirmed:
> >Not in a match, for sure. :)
>
> Nor in a single game or anything else where there is a full point doled out
> for
> each game!
>
OK, to be clear, you can only win by half a point when playing in a
tournament against more than one player. My half-point comment was
rhetorical... ;-)
Henri
Mhoulsby
>Date: 18/11/03 12:29 GMT Standard Time
>Message-id: <arseno-E3AE31....@news.ulaval.ca>
Not so. It proves that Fritz plays well in open positions with lots of tactics,
and is clueless in closed positions. It proves that if Gonzo had adopted an
anticomputer strategy in game one, he would have won game one (will he never
learn?) Of course, we already knew that, and we struggle to beat it even on its
lowest settings (well, I do, anyway).
Mark
Alberich
says...
moves played to Deep Fritz on Sunday." Well, here's Crafty's analysis as
I turned on infinite analysis and had Crafty 19.03 look at the position.
Here's it's analysis:
Kasparov-FritzX3D
r1bqk2r/1p1nbppp/1Pp2n2/p1Pp4/Q2Pp3/B1N1P2P/P2N1PP1/R3KB1R b KQkq - 0 1
Analysis by Crafty 19.03:
1... O-O 2. Nb3 Ne8 3. Nxa5 f5! 4. Bb2 Nxb6 5. cxb6 Qxb6 6. Nc4 dxc4 7.
Bxc4+ Kh8 8. Qxa8 Qxb2 9. O-O Nd6 10. Na4 Qb4 11. Bb3 f4 12. Qa7 Nb5 13.
Qb6 Bg5
Y'know, I think Crafty is a better engine than Fritz. I'm going to keep
it as my main engine to play against instead of Fritz in the Chessbase
GUI. I'll return to Fritz when the programmers get their heads out of
the sand and improve Fritz' ability to play in closed complex positions.
Ed Seedhouse
<dav...@chiark.greenend.org.uk> wrote:
>henri Arsenault <ars...@nospamhere.phy.ulaval.ca> wrote:
>> Whatever the result is tomorrow, and whatever conclusion is drawn from
>> the result, the conclusion will be based on a single game, and therefore
>> of little statistical value.
>
>No, the conclusion will be based on four games. To claim that it is based
>on just one is to claim that the result of the first three games does not
>influence the outcome of the match.
Dave is of course absolutely correct, here. Still, a sample of four
games is pretty small and even if one side won four straight the error
range would be fairly broad. In the event neither side can win by
more than one point and I think that if you crunch the numbers you
won't be able to prove much one way or the other based on that kind of
result.
Mhoulsby
>Date: 18/11/03 14:04 GMT Standard Time
>Message-id: <tj9krv8hcacuk636j...@4ax.com>
Well, it seems to me that Fritz, and engines generally, have a history of
playing increasingly badly the more closed positions become.
They play badly in endgames which are not tablebases, too.
Therefore, four games certainly provides more evidence to add to the weight of
evidence already in existence, which demonstrate that these shortcomings in
engines' chessplaying remain to be overcome.
Robert Hyatt
> Analysis by Crafty 19.03:
Let me point out one flaw. Just because Crafty has a bit better
"understanding" (if that is a reasonable word) of this kind of blocked
position does _not_ mean it is better overall. Crafty has grown up in
a different "world" than Fritz. IE the Internet Chess Club and various
other servers. There it meets master/IM/GM players that are more than
satisfied with a draw, because Crafty's rating is usually higher than
theirs and they win "points" with a draw. I've been combatting this
particular kind of chess for several years. But if you don't play and
tune on ICC (or another server) then you won't see the "draw-masters"
and the need to combat their blocking the position into a dead draw.
Crafty is not bad at this part of the game, but it is a two-edged
sword.
Robert Hyatt
Of course, you need to look at the margin of error for a 4-game match
against players that are reasonably close (within 200 points) in playing
skill. It is quite large.
Which means the worse player could win the match a significant number
of times.
> --
> A fool can beat any foolproof system.
> 14th saying of Bernard
--
Robert Hyatt
>>From: Ed Seedhouse eseed...@510shaw.ca
>>Date: 18/11/03 14:04 GMT Standard Time
>>Message-id: <tj9krv8hcacuk636j...@4ax.com>
>>
>>On 17 Nov 2003 23:39:24 +0000 (GMT), David Richerby
>><dav...@chiark.greenend.org.uk> wrote:
>>
>>>henri Arsenault <ars...@nospamhere.phy.ulaval.ca> wrote:
>>>> Whatever the result is tomorrow, and whatever conclusion is drawn from
>>>> the result, the conclusion will be based on a single game, and therefore
>>>> of little statistical value.
>>>
>>>No, the conclusion will be based on four games. To claim that it is based
>>>on just one is to claim that the result of the first three games does not
>>>influence the outcome of the match.
>>
>>Dave is of course absolutely correct, here. Still, a sample of four
>>games is pretty small and even if one side won four straight the error
>>range would be fairly broad. In the event neither side can win by
>>more than one point and I think that if you crunch the numbers you
>>won't be able to prove much one way or the other based on that kind of
>>result.
>>
> Well, it seems to me that Fritz, and engines generally, have a history of
> playing increasingly badly the more closed positions become.
> They play badly in endgames which are not tablebases, too.
I can only address my program here, but this is definitely not true in the
case of Crafty. 7-8 years ago I dreaded to see Crafty enter an endgame when
playing a GM, even if it was a pawn up. Today I don't fear endgames at all
and I don't see _any_ evidence that suggests that crafty (specifically) loses
more endgames than it does tactically open middlegames.
I spent a lot of time adding specific endgame knowledge dealing with
passed pawns, distant passed pawns, majorities, distant majorities,
and lots of special-case stuff such as bishop + wrong rook pawn(s) draws
and the like. And a critical piece of data that in king and pawn endings,
if one side has two connected passers and the other side has two split
(isolated on opposite sides of the board) passers, the connected passers
do _not_ win.
Crafty is nowhere near perfect. But it plays the endgame pretty well
anyway. Ask someone like Roman that has tried it in hundreds of games
on ICC. I can remember the Cray Blitz days where we knew full well that
we had our share of weaknesses, particularly in endgames. I chose to
eliminate as many of those as possible during the development of Crafty.
> Therefore, four games certainly provides more evidence to add to the weight of
> evidence already in existence, which demonstrate that these shortcomings in
> engines' chessplaying remain to be overcome.
--
David Richerby
No! If the score is even after three games, that suggests (albeit with a
high margin of error) that the playing strengths are roughly equal. If
the fourth game is drawn, this suggests (with a slightly lower margin of
error) that the playing strengths are equal. Likewise, if either player
wins game four, then it suggests that one of the players is slightly
better, again with a fairly high margin of error.
Dave.
--
David Richerby Evil Whisky (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ single-malt whisky but it's genuinely
evil!
Robert Hyatt
> henri Arsenault <ars...@nospamhere.phy.ulaval.ca> wrote:
>> David Richerby <dav...@chiark.greenend.org.uk> wrote:
>>> henri Arsenault <ars...@nospamhere.phy.ulaval.ca> wrote:
>>>> Whatever the result is tomorrow, and whatever conclusion is drawn
>>>> from the result, the conclusion will be based on a single game, and
>>>> therefore of little statistical value.
>>>
>>> No, the conclusion will be based on four games. To claim that it is
>>> based on just one is to claim that the result of the first three games
>>> does not influence the outcome of the match.
>>
>> No, it is to conclude that since the score is even after 3 games, these
>> games have proven nothing about relative strength, and they will not
>> have proven anything more by adding a fourth game.
> No! If the score is even after three games, that suggests (albeit with a
> high margin of error) that the playing strengths are roughly equal. If
> the fourth game is drawn, this suggests (with a slightly lower margin of
> error) that the playing strengths are equal. Likewise, if either player
> wins game four, then it suggests that one of the players is slightly
> better, again with a fairly high margin of error.
> Dave.
Yes, but let's be very precise here:
1. three draws and then a win says something totally different
from
2. a single win.
The TPR (performance rating) will be different for the two cases.
The error bar will be different.
With either you could conclude that the winning side is better. case
1 produces a more reliable estimate of how much better, and the value
will be smaller since it factors in the three draws (or 1 draw, 1 win,
1 loss as in the real match, that's the same as three draws statistically
here).
> --
> David Richerby Evil Whisky (TM): it's like a
> www.chiark.greenend.org.uk/~davidr/ single-malt whisky but it's genuinely
> evil!
--
Alberich
You are aware, of course, that I'm using a WinUCI adapter to even USE
Crafty in the Chessbase GUI, don't you? Although Fritz comes with an
early version of Crafty...I'm using one of the latest versions of the
program in the Chessbase GUI. But you have to explain to me why your
program fails against Winboard engine Amy almost all the time. I've had
engine matches between them on my computer and Amy constantly whips
Crafty. Please try a winboard match between these two. Trust me you'd be
surprised.
Mhoulsby
>Date: 18/11/03 14:47 GMT Standard Time
>Message-id: <bpdbdo$dh1$6...@juniper.cis.uab.edu>
Indeed. I was not thinking of Crafty at all.
'Mao: I spoke generally. The line
We take now is a paradox...'
--Alice Goodman (Nixon In China, music John Adams, 1987)
<snip>
Mark
Ed Seedhouse
<hy...@crafty.cis.uab.edu> wrote:
>Yes, but let's be very precise here:
>1. three draws and then a win says something totally different
>from
>2. a single win.
>The TPR (performance rating) will be different for the two cases.
>The error bar will be different.
All perfectly true, so far.
>With either you could conclude that the winning side is better.
Here I must disagree, at least if you are talking about a rational
conclusion. The standard deviation of performance in a single game is
300 rating points. The standard error for four games will, therefore,
be 300/sqrt(4) or 150 points. Unless the performance of the winner is
substantially in excess of one standard deviation above the
performance of the loser it will not be reasonable to claim with any
level of certainty that the winner is "better" than the other based on
such a small sample.
Of course one can make any conclusion one likes on any evidence one
likes, or no evidence at all for that matter, but will it be a
reasonable conclusion? If it is not based on good evidence I don't
think it will be.
Ed Seedhouse
wrote:
>>Dave is of course absolutely correct, here. Still, a sample of four
>>games is pretty small and even if one side won four straight the error
>>range would be fairly broad. In the event neither side can win by
>>more than one point and I think that if you crunch the numbers you
>>won't be able to prove much one way or the other based on that kind of
>>result.
>Therefore, four games certainly provides more evidence to add to the weight of
>evidence already in existence, which demonstrate that these shortcomings in
>engines' chessplaying remain to be overcome.
But all this has nothing whatsoever to do with how strong the computer
is. Of course it has weaknesses. But all we have to go by when
measuring strength of play is performance. And over four games the
evidence of performance is probably not enough to draw meaningful
conclusions about the comparative strength of the two players unless
the score it totally lopsided, which it certainly wasn't.
Mhoulsby
>Date: 19/11/03 00:57 GMT Standard Time
>Message-id: <pmflrvki2p2e5tcd2...@4ax.com>
Again, I suspect that it's important not to *disregard* all of the games which
*earlier versions* of Fritz have played against Garry Kimovich, especially, but
also against all GMs it has played. Then take into account games which GKK has
played against those same GMs. That makes the sample size bigger, maybe even
critically significant.
Mhoulsby
>Date: 19/11/03 01:00 GMT Standard Time
>Message-id: <71glrv0m56kfp1ku0...@4ax.com>
>
>On 18 Nov 2003 14:22:55 GMT, mhou...@aol.com-remove- (Mhoulsby)
>wrote:
>
>>>Dave is of course absolutely correct, here. Still, a sample of four
>>>games is pretty small and even if one side won four straight the error
>>>range would be fairly broad. In the event neither side can win by
>>>more than one point and I think that if you crunch the numbers you
>>>won't be able to prove much one way or the other based on that kind of
>>>result.
>
>>Therefore, four games certainly provides more evidence to add to the weight
>of
>>evidence already in existence, which demonstrate that these shortcomings in
>>engines' chessplaying remain to be overcome.
>
>But all this has nothing whatsoever to do with how strong the computer
>is.
Now you're just being stupid again. Clearly it doesn't have *nothing
whatsoever* to do with how strong the computer is. If one takes into account
Fritz' past performance against GMs and GM strength non-human opponents, the
number of games begins to become increasingly significant, perhaps even
critically significant. Remember, even Fritz3 caused Garry Kimovich some
problems, once upon a time.
> Of course it has weaknesses. But all we have to go by when
>measuring strength of play is performance.
...and analysis, yes.
And over four games the
>evidence of performance is probably not enough to draw meaningful
>conclusions about the comparative strength of the two players unless
>the score it totally lopsided, which it certainly wasn't.
>
Even that depends upon your point-of-view.
It seems to me that Garry Kimovich does far less well against computers in
general than is necessary. If he'd played an *exclusively* anti-computer
strategy in 1996, there's good reason to suspect that he would not have lost a
single game.
Equally, if he had played in game 1 of the match against X3D Fritz the way he
played in game 3, which of us can put his (or her) hand on his (or her) heart
and say that he would not have scored +2-0=2?
You're big on certainties, Ed. Sometimes, however, *like me*, the things about
which you seem certain turn out to be wrong.
Ed Seedhouse
wrote:
>>Here I must disagree, at least if you are talking about a rational
>>conclusion. The standard deviation of performance in a single game is
>>300 rating points. The standard error for four games will, therefore,
>>be 300/sqrt(4) or 150 points. Unless the performance of the winner is
>>substantially in excess of one standard deviation above the
>>performance of the loser it will not be reasonable to claim with any
>>level of certainty that the winner is "better" than the other based on
>>such a small sample.
>Again, I suspect that it's important not to *disregard* all of the games which
>*earlier versions* of Fritz have played against Garry Kimovich, especially, but
>also against all GMs it has played. Then take into account games which GKK has
>played against those same GMs. That makes the sample size bigger, maybe even
>critically significant.
But this is merely shifting ground. The original question under
discussion was how much evidence this single match will provide about
the relative strengths of the two players. That's certainly what I
was speaking to.
If we want to take into account all of Kasparov's games against
computers we can possibly draw certain conclusions about Kasparov's
strength relative to the computers he has played. But it seems rather
silly to draw general coclusions about the strength of a particular
computer based on Kasparov's games against *other* computers.
Mhoulsby
>Date: 19/11/03 01:44 GMT Standard Time
>Message-id: <7jilrvsu5amr3dej2...@4ax.com>
>
>On 19 Nov 2003 01:02:33 GMT, mhou...@aol.com-remove- (Mhoulsby)
>wrote:
>
>
>>>Here I must disagree, at least if you are talking about a rational
>>>conclusion. The standard deviation of performance in a single game is
>>>300 rating points. The standard error for four games will, therefore,
>>>be 300/sqrt(4) or 150 points. Unless the performance of the winner is
>>>substantially in excess of one standard deviation above the
>>>performance of the loser it will not be reasonable to claim with any
>>>level of certainty that the winner is "better" than the other based on
>>>such a small sample.
>
>>Again, I suspect that it's important not to *disregard* all of the games
>which
>>*earlier versions* of Fritz have played against Garry Kimovich, especially,
>but
>>also against all GMs it has played. Then take into account games which GKK
>has
>>played against those same GMs. That makes the sample size bigger, maybe even
>>critically significant.
>
>But this is merely shifting ground. The original question under
>discussion was how much evidence this single match will provide about
>the relative strengths of the two players. That's certainly what I
>was speaking to.
>
Yes, you were, but *my* point was that in order to examine *this match* in its
*proper context*, one should probably examine *all* of the younger Garry
Kimovich's encounters with the younger Fritz, as well.
>If we want to take into account all of Kasparov's games against
>computers we can possibly draw certain conclusions about Kasparov's
>strength relative to the computers he has played. But it seems rather
>silly to draw general coclusions about the strength of a particular
>computer based on Kasparov's games against *other* computers.
>
Agreed, which is why I specifically emphasised: "*earlier* versions of Fritz".
My having cited the DB fiasco was an aside.
Ed Seedhouse
wrote:
>Now you're just being stupid again.
Ah well, that comprehensively disproves everyting I say!! Your
brilliant argument has totally destroyed me. I'm so bitter about your
proof of my stupidity I can think of nothing else to do but stick you
into my ignore file from now on.
Mhoulsby
>Date: 19/11/03 02:11 GMT Standard Time
>Message-id: <b6klrvslfs1bjr7rt...@4ax.com>
How many times have you claimed that you would do that?
You're an idiot troll, Ed, nothing more.
Bernardz
dav...@chiark.greenend.org.uk says...
> Bernardz <Berna...@REMOVEhotmail.com> wrote:
> > dav...@chiark.greenend.org.uk says...
> >> No, the conclusion will be based on four games. To claim that it is
> >> based on just one is to claim that the result of the first three
> >> games does not influence the outcome of the match.
> >
> > Disagree. You have to go by the score.
>
> You don't appear to be disagreeing with anything I said; in fact, you seem
> to be agreeing with me. Please try to follow up to the right post.
Point taken.
>
>
> > If Fritz wins tomorrow, it won two games, drew one and lost one. So we
> > would have to conclude that it was better.
>
> You really can't conclude much from a four-game match. It's too
> susceptible to external factors. Imagine Kasparov turns up to today's
> game with a blistering headache and loses: in a twenty-game match, losing
> one game because of a headache would have very little effect; in a
> four-game match, there's no time to come back. So, if Fritz wins this
> match, you have a little bit of evidence that Fritz plays better chess,
> but not very much. You need a longer match to be able to say with
> reasonable certainty that one player is better than another.
Then why do it? All we can do is access the result on the score. As it
was the results a draw seem quite a fair result.
>
>
> Dave.
David Richerby
>> able to say with reasonable certainty that one player is better than
>> another.
>
> Then why do it?
To entertain people, to produce interesting chess, to make money, to
publicize Fritz, Kasparov and the X3D people. I'd say it was pretty
successful in doing all of those things.
Dave.
--
David Richerby Evil Mouldy Gerbil (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ children's pet but it's starting to
grow mushrooms and genuinely evil!